UMN Seasonal Crime Rates CMSC320 Final Project

Introduction:

This is a tutorial to teach the entire process of data science and how data curation, analysis, machine learning, and hypothesis testing can be used in any field of study. Data science is a mix between hacking skills, math and statistics knowledge, and substantive expertise. The specific field of study I chose to show this pipeline is Minneapolis Incidents & Crime. This shows public-facing service requests taken through the Lagan CRM system, and police incidents in Minneapolis Minnesota respectively. The reason I chose to use this topic is because I will be living in Minneapolis this summer for 10 weeks for my summer internship, so I figured it was a good idea to prove with data science that it’s a safe place to live during the summer.

Data Curation, Parsing, and Management;

Setup for the R Markdown file.

knitr::opts_chunk$set(echo = TRUE)

Importing the libraries we will use for this tutorial.

library(tidyverse)

## -- Attaching packages -------------------------------------------------------------------------------------------------- tidyverse 1.2.1 --

## v ggplot2 2.2.1     v purrr   0.2.4
## v tibble  1.4.2     v dplyr   0.7.4
## v tidyr   0.7.2     v stringr 1.2.0
## v readr   1.1.1     v forcats 0.2.0

## -- Conflicts ----------------------------------------------------------------------------------------------------- tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag()    masks stats::lag()

library(lubridate) 

## 
## Attaching package: 'lubridate'

## The following object is masked from 'package:base':
## 
##     date

library(ggplot2)
library(dplyr)
library(leaflet)

## Warning: package 'leaflet' was built under R version 3.4.4

library(viridis)

## Warning: package 'viridis' was built under R version 3.4.4

## Loading required package: viridisLite

## Warning: package 'viridisLite' was built under R version 3.4.4

library(ISLR)

## Warning: package 'ISLR' was built under R version 3.4.4

library(tree)

## Warning: package 'tree' was built under R version 3.4.4

Initializing the datasets. They can be found online at: https://www.kaggle.com/mrisdal/minneapolis-incidents-crime

crimes <- read_csv("crimes.csv")

## Parsed with column specification:
## cols(
##   .default = col_character(),
##   controlnbr = col_integer(),
##   Precinct = col_integer(),
##   ReportedDate = col_datetime(format = ""),
##   BeginDate = col_datetime(format = ""),
##   Time = col_time(format = ""),
##   UCRCode = col_integer(),
##   EnteredDate = col_datetime(format = ""),
##   Long = col_double(),
##   Lat = col_double(),
##   x = col_double(),
##   y = col_double(),
##   lastchanged = col_datetime(format = ""),
##   LastUpdateDate = col_datetime(format = "")
## )

## See spec(...) for full column specifications.

incidents <- read_csv("incidents.csv")

## Parsed with column specification:
## cols(
##   X = col_double(),
##   Y = col_double(),
##   CASEID = col_double(),
##   ENQUIRYTYPEID = col_integer(),
##   SUBJECTNAME = col_character(),
##   REASONNAME = col_character(),
##   TYPENAME = col_character(),
##   TITLE = col_character(),
##   OPENEDDATETIME = col_datetime(format = ""),
##   CASESTATUS = col_integer(),
##   CLOSEDDATETIME = col_datetime(format = ""),
##   XCOORD = col_double(),
##   YCOORD = col_double(),
##   LastUpdateDate = col_datetime(format = ""),
##   YEAR = col_integer()
## )

Let’s take a look at the two datasets.

#police incidents in Minneapolis, Minnesota from 2010 to 2016
head(crimes)

## # A tibble: 6 x 20
##   publicaddress        controlnbr CCN         Precinct ReportedDate       
##   <chr>                     <int> <chr>          <int> <dttm>             
## 1 0029XX Chicago AV S        2001 MP 2010 04~        3 2010-02-22 13:40:00
## 2 0049XX Queen AV N          2002 MP 2010 99~        4 2010-02-22 13:38:15
## 3 0003XX 16 AV SE            2003 MP 2010 99~        2 2010-02-22 13:40:50
## 4 0016XX University A~       2004 MP 2010 04~        2 2010-02-22 14:15:00
## 5 00003X 9 ST S              2005 MP 2010 04~        1 2010-02-22 15:00:00
## 6 0003XX Washington A~       2006 MP 2010 04~        2 2010-02-22 15:17:00
## # ... with 15 more variables: BeginDate <dttm>, Time <time>,
## #   Offense <chr>, Description <chr>, UCRCode <int>, EnteredDate <dttm>,
## #   Long <dbl>, Lat <dbl>, x <dbl>, y <dbl>, Neighborhood <chr>,
## #   lastchanged <dttm>, LastUpdateDate <dttm>, OBJECTID <chr>,
## #   ESRI_OID <chr>

#public-facing service requests that were taken by 311 contact center agents, or through online and mobile applications, and entered the Lagan CRM system from 2011 to 2015
head(incidents)

## # A tibble: 6 x 15
##       X     Y   CASEID ENQUIRYTYPEID SUBJECTNAME REASONNAME TYPENAME TITLE
##   <dbl> <dbl>    <dbl>         <int> <chr>       <chr>      <chr>    <chr>
## 1 -93.2  44.9  1.01e11          1001 Public Saf~ Graffiti ~ Graffit~ Graf~
## 2 -93.3  45.0  1.01e11          1002 Animal Rel~ Animal Co~ Animal ~ Anim~
## 3 -93.3  45.0  1.01e11          1003 Property    Private P~ Residen~ Resi~
## 4 -93.3  45.0  1.01e11          1004 Animal Rel~ Animal Co~ Animal ~ Anim~
## 5 -93.3  45.0  1.01e11          1005 Vehicles a~ Traffic C~ Parking~ <NA> 
## 6 -93.3  45.0  1.01e11          1006 Vehicles a~ Traffic C~ Abandon~ <NA> 
## # ... with 7 more variables: OPENEDDATETIME <dttm>, CASESTATUS <int>,
## #   CLOSEDDATETIME <dttm>, XCOORD <dbl>, YCOORD <dbl>,
## #   LastUpdateDate <dttm>, YEAR <int>

Let’s take a deeper look into the crime table as I’ll be focusing my efforts on that one. R simply opened up the csv (comma separated value) file with the function read_csv and then displayed the first 6 rows of the table in an elegant fashion. It displays different types of data like publicaddress, Time, and Offense which are attributes heading each columns of data for a particular crime which is said to be an entity.

One thing I want for my experiment is to know just the year and month of the crime since I’m considering yearly trends and seasonal safety specifically during the summer months. So I am going to tidy up the data and alter the data frame by adding columns that represents year and month of the crime as well as hour from ReportedDate attribute.

#adding a new column and using format to extract the year and month of the crime
crimes$year_month <- format(as.Date(crimes$ReportedDate), "%Y-%m")
crimes$year <- format(as.Date(crimes$ReportedDate), "%Y")
crimes$month <- format(as.Date(crimes$ReportedDate), "%m")
crimes$hour <- factor(substr(crimes$Time, 1, 2))
#notice the new year and month columns at the end
head(crimes)

## # A tibble: 6 x 24
##   publicaddress        controlnbr CCN         Precinct ReportedDate       
##   <chr>                     <int> <chr>          <int> <dttm>             
## 1 0029XX Chicago AV S        2001 MP 2010 04~        3 2010-02-22 13:40:00
## 2 0049XX Queen AV N          2002 MP 2010 99~        4 2010-02-22 13:38:15
## 3 0003XX 16 AV SE            2003 MP 2010 99~        2 2010-02-22 13:40:50
## 4 0016XX University A~       2004 MP 2010 04~        2 2010-02-22 14:15:00
## 5 00003X 9 ST S              2005 MP 2010 04~        1 2010-02-22 15:00:00
## 6 0003XX Washington A~       2006 MP 2010 04~        2 2010-02-22 15:17:00
## # ... with 19 more variables: BeginDate <dttm>, Time <time>,
## #   Offense <chr>, Description <chr>, UCRCode <int>, EnteredDate <dttm>,
## #   Long <dbl>, Lat <dbl>, x <dbl>, y <dbl>, Neighborhood <chr>,
## #   lastchanged <dttm>, LastUpdateDate <dttm>, OBJECTID <chr>,
## #   ESRI_OID <chr>, year_month <chr>, year <chr>, month <chr>, hour <fct>

crimes

## # A tibble: 136,121 x 24
##    publicaddress        controlnbr CCN        Precinct ReportedDate       
##    <chr>                     <int> <chr>         <int> <dttm>             
##  1 0029XX Chicago AV S        2001 MP 2010 0~        3 2010-02-22 13:40:00
##  2 0049XX Queen AV N          2002 MP 2010 9~        4 2010-02-22 13:38:15
##  3 0003XX 16 AV SE            2003 MP 2010 9~        2 2010-02-22 13:40:50
##  4 0016XX University A~       2004 MP 2010 0~        2 2010-02-22 14:15:00
##  5 00003X 9 ST S              2005 MP 2010 0~        1 2010-02-22 15:00:00
##  6 0003XX Washington A~       2006 MP 2010 0~        2 2010-02-22 15:17:00
##  7 0055XX 28 AV S             2007 MP 2010 0~        3 2010-02-22 15:47:00
##  8 0008XX 5 ST SE             2008 MP 2010 0~        2 2010-02-22 16:00:00
##  9 0051XX Camden AV N         2009 MP 2010 0~        4 2010-02-22 16:35:00
## 10 0023XX Lake ST E           2010 MP 2010 0~        3 2010-02-22 17:22:00
## # ... with 136,111 more rows, and 19 more variables: BeginDate <dttm>,
## #   Time <time>, Offense <chr>, Description <chr>, UCRCode <int>,
## #   EnteredDate <dttm>, Long <dbl>, Lat <dbl>, x <dbl>, y <dbl>,
## #   Neighborhood <chr>, lastchanged <dttm>, LastUpdateDate <dttm>,
## #   OBJECTID <chr>, ESRI_OID <chr>, year_month <chr>, year <chr>,
## #   month <chr>, hour <fct>

We can also use the previously imported library leaflet to create an interactive map of Minneapolis.

#here is a basic interactive map of the city
min_map <- leaflet(crimes) %>%
  addTiles() %>%
  setView(lat = 44.9778, lng = -93.2650, zoom = 11)
min_map

#now we can add circle markers for each crime
map_circle <- leaflet(crimes) %>%
  addTiles() %>%
  setView(lat = 44.9778, lng = -93.2650, zoom = 11) %>%
  addCircleMarkers(data = crimes, lng = ~ Long, lat = ~ Lat, radius = 2, clusterOptions = markerClusterOptions()) 
map_circle

#creating a dataframe for number of crimes for each neighborhood with the average longitude and latitude
crimes_num <- crimes %>%
  group_by(Neighborhood) %>%
  summarize(num_crimes = n(), Long = mean(Long), Lat = mean(Lat))
crimes_num

## # A tibble: 88 x 4
##    Neighborhood         num_crimes  Long   Lat
##    <chr>                     <int> <dbl> <dbl>
##  1 ARMATAGE                    455 -93.3  44.9
##  2 AUDUBON PARK               1050 -93.2  45.0
##  3 BANCROFT                    800 -93.3  44.9
##  4 BELTRAMI                    384 -93.2  45.0
##  5 BOTTINEAU                   636 -93.3  45.0
##  6 BRYANT                      614 -93.3  44.9
##  7 BRYN - MAWR                 627 -93.3  45.0
##  8 CAMDEN INDUSTRIAL           151 -93.3  45.0
##  9 CARAG                      1776 -93.3  44.9
## 10 CEDAR - ISLES - DEAN        815 -93.3  45.0
## # ... with 78 more rows

#color function to show severity of total crime count
getColor <- function(crime) {
  sapply(crime$num_crimes, function(num_crimes) {
  if(num_crimes <= 1000) {
    "green"
  } else if(num_crimes <= 5000) {
    "orange"
  } else {
    "red"
  } })
}

#function to create cool looking icons
icons <- awesomeIcons(
  icon = 'ios-close',
  iconColor = 'black',
  library = 'ion',
  markerColor = getColor(crimes_num)
)

#finally let's show the total number of crimes for each neighborhood with markers for each neighborhood
map_avg <- leaflet(crimes_num) %>%
  addTiles() %>%
  setView(lat = 44.9778, lng = -93.2650, zoom = 11) %>%
  addAwesomeMarkers(~Long, ~Lat, icon = icons, label = ~as.character(num_crimes))
#green markers are safe, orange are dangerous, and red are seriously dangerous (downtown)
map_avg

Exploratory Data Analysis;

Now let’s analyze the data with a graph. There are three steps to this: 1) The data that goes into the plot, which in this case is our crime data frame of entities and attributes. 2) The mapping between the attributes and the aesthetic graphical qualities. 3) The geometric display of these qualities in terms of the graph.

I will use some simple R code that takes care of each of these steps.

#new crime plot plot object, this is the data from 1)
  crime_plot <- crimes %>%
  #filtering to get more recent data
  filter(year > 2011) %>%
  #grouping by year and month to see trends
  group_by(year_month) %>%
  #creating a new temporary variable for the y axis
  summarize(num_crimes = n()) %>%
  #to get rid of one outlier and tidy up the graph
  filter(num_crimes > 1000) %>%
  #this is the mapping from 2)
  ggplot(aes(x = year_month, y = num_crimes)) +
    #this is the geometric display from 3) and all the optional aesthetics to make it presentable
    geom_point() +
    xlab("Year-Month") +
    ylab("Number of Crimes") +
    theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
    ggtitle("Number of Crimes Each Year") 
crime_plot

Fairly successful plot to get us started. I start by filtering the year to look at 2012 and beyond. I then group by the year and month to collect counts for each month of crimes. I then make a new temporary attribute for the y axis and filter out the one outlier month since it made the graph overly tall. I then plot the data with a scatterplot and tidy it up with titles and making the x axis labels sideways so they’re less ugly.

Next, I want to limit the Neighborhood to University of Minnesota since that’s where I will be living. I will also group by month to make things more clear using my new year attribute.

  crime_plot <- crimes %>%
  filter(year > 2011) %>%
  #limiting to crimes that happen at UMN
  filter(Neighborhood == "UNIVERSITY OF MINNESOTA") %>%
  group_by(month) %>%
  summarize(num_crimes = n()) %>%
  ggplot(aes(x = month, y = num_crimes)) +
    geom_point() +
    xlab("Month") +
    ylab("Number of Crimes") +
    ggtitle("number of Crimes Each Year and Month for University of Minnesota")  
crime_plot

Important things to pay attention to with graph for exploratory data analysis are central trends, spread, skew, and outliers. Let’s classify each,

Central trends: The graph centers around about 35 crimes. This can be thought of as the mean or average. Spread: This can be thought of the variance of the data which is pretty high, having some at about 20 and some as high as 50. Skew: It is positively skewed meaning the right half of the data after June is higher. Outliers: In the first graph, I got rid of an outlier but in the second I would say there’s an outlier in December probably because of the cold weather.

Using the method summary, we can backup my visual observations with a statistics markdown of the exploratory data analysis.

crimes_sum <- summary(
    crimes %>%
    filter(year > 2011) %>%
    filter(Neighborhood == "UNIVERSITY OF MINNESOTA") %>%
    group_by(month) %>%
    summarize(num_crimes = n()) 
  )
crimes_sum

##     month             num_crimes   
##  Length:12          Min.   :21.00  
##  Class :character   1st Qu.:30.75  
##  Mode  :character   Median :37.00  
##                     Mean   :36.25  
##                     3rd Qu.:42.25  
##                     Max.   :47.00

Hypothesis Testing;

We can use a basic inference method called the inverse problem. Essentially we have a population of all the crimes in Minneapolis. We want to make a probabilistic model to generate data. Our inverse problem is inferring the parameter of Minneapolis being safe compared to America overall. Also showing that the three summer months are safer.

Hypothesis testing is a statistics method that tests a scientific hypothesis against a population normal. I got data from nationmaster.com and it shows that there are 41.29 crimes per 1000 people in the United States in 2002. We are going to test this against our sample of data from Minneapolis. Our hypotheses are,

(null), \[ H_0: p \leq \frac{41.29}{1000} = 0.04129 \] (alternative), \[ H_1: p > 0.04129 \]

Next we need to calculate the average number of crimes in Minneapolis for each year relative to the population that I got from Google which is 413,651 people in 2016.

population <- 413651
crimes_avg <- crimes %>%
  group_by(year) %>%
  summarize(num_crimes = n())
#num_crimes being the number of crimes for each year from our data
crimes_avg

## # A tibble: 7 x 2
##   year  num_crimes
##   <chr>      <int>
## 1 2010       20339
## 2 2011       21629
## 3 2012       21930
## 4 2013       21857
## 5 2014       21526
## 6 2015       19995
## 7 2016        8845

#average number of crimes each year
average <- mean(crimes_avg$num_crimes)
average

## [1] 19445.86

We assume that our null hypothesis is true unless there is compelling evidence to reject it. The event is statistically significant if it has 5% or less probability of happening. Here, this 5% is called the significance level of the test, denoted alpha, alpha = 0.05.

#expected crime rate from our sample with the given population in Minneapolis 
expect <- average / population
expect

## [1] 0.0470103

Now that we have a phat which is our proposed p (expect), we can create a confidence interval to see if we can reject the null hypothesis or not.

#estimate for our proposed phat
get_estimate <- function(n, p = 0.041) mean(sample(c(0 , 1), size = n, replace = TRUE, prob = c( 1 - p, p)))
#new data frame for our confidence interval and hypothesis testing
tab <- data.frame(sample_size = c(population)) %>%
  #column for phat which is our proposed p estimate
  mutate(phat = sapply(sample_size, get_estimate)) %>%
  #standard error for our estimate
  mutate(se = sqrt(expect * (1 - expect)) / sqrt(population)) %>%
  #the lower left tail of our graph
  mutate(lower = expect + qnorm(.05 / 2, sd = se)) %>%
  #the upper right tail of our graph
  mutate(upper = expect + -qnorm(.05 / 2, sd = se)) %>%
  #the p_value being the pnorm which is the z value for the normal distribution
  mutate(p_value = 1 - pnorm(expect, mean = 0.041, sd = se))
tab

##   sample_size       phat           se      lower      upper p_value
## 1      413651 0.04067922 0.0003290967 0.04636528 0.04765532       0

Our hypothesis testing gives us a phat which again is our estimate for p and it turns out to be 0.41 which is roughly equal to our null hypothesis. Since the null is being less than or equal to 0.041, our 0.041 is within 0.05 of our null, so we definitely can’t reject our null hypothesis.

Machine Learning to Provide Analysis;

First I am going to analyze crime over time throughout the day and see if the general number of crimes is increasing or decreasing in Minneapolis throughout the day to see when it’s safest for me to be outside this summer.

#let's average the number of crimes by each month and year similar to how we did with our graph earlier
crimes_avg <- crimes %>%
  group_by(Time) %>%
  summarize(num_crimes = n()) %>%
  #to get rid of outliers 
  filter(num_crimes < 500)
crimes_avg

## # A tibble: 1,396 x 2
##    Time   num_crimes
##    <time>      <int>
##  1 02'00"         37
##  2 03'00"         19
##  3 04'00"         18
##  4 05'00"        132
##  5 06'00"         17
##  6 07'00"         20
##  7 08'00"         25
##  8 09'00"         25
##  9 10'00"        115
## 10 11'00"         18
## # ... with 1,386 more rows

We can use a machine learning process called linear regression that is strong, basic, and elegant to use when analyzing data. We can use a statistical model with our data set to predict outcomes for future crime numbers.

With the model we’re trying to create, we have a continuous numerical variable of Time and a numerical variable num_crimes. We go ahead and assume that our prediction has a relationship that can be represented as a linear function like:

\[ num\_crimes = \beta_0 + \beta_1 \cdot Time \]

We have this data for our sample, but we want to be able to predict the population which would be the city of Minneapolis since our data is a small subset with a limited Time frame. So we need to find values for beta_0 and beta_1 that minimize loss with a RSS which stands for residual sum of squares. Let us put this together with some code and plot it using a linear regression model.

crimes_avg %>%
  #basic dot plot with crimes for each month of every year
  ggplot(aes(x = Time, y = num_crimes)) + 
  geom_point() +
  labs(x="Time Crime Occurs") + 
  labs(y="Numbers of Crimes During this Time") + 
  ggtitle("Number of Crimes for Each Unique Time of Day") +
  #lm stands for linear model, so our method is linear regression. This is a geometric representation of what we're trying to show
  geom_smooth(method = lm) 

This scatterplot shows the fluctuation and variation amongst crimes with most Times being low on crimes but some having huge spikes. The saturation and height of the number of crimes increases throughout the day as shown by the regression lines which basically tries to best fit the date. We removed Times where there were 500 or more crimes since those make the line jump up more to make it fit less of the majority of Times throughout the day.

We can show how this line is made by doing some more code.

model <- lm(Time ~ num_crimes, data = crimes_avg)
model

## 
## Call:
## lm(formula = Time ~ num_crimes, data = crimes_avg)
## 
## Coefficients:
## (Intercept)   num_crimes  
##    40172.49        91.49

so our beta_0 = 40172.49 and our beta_1 = 91.49. This makes sense because there are the most crimes when Time = 0 because that’s at midnight. Also the way crimes are reported with this dataset, a lot seem to end up falling on this time.

Speaking of regression, another tool we can use to show predictions based on data is regression trees. They are essentially paths that you follow to see the average prediction based on what bucket our date is in.

crimes_hourly <- crimes %>%
  group_by(hour) %>%
  summarize(num_crimes = n()) 

tree <- tree(num_crimes ~ hour, data = crimes_hourly)
plot(tree)
text(tree, pretty = 0, cex = 1)

This data is the same as the other graph, except this one is easier to read and shows it in a different way. As an example, if we want crime numbers between 8:00am and 12:00pm we go to the leaf that has 4052 crimes because it’s between those two numbers.

Curation of a Message or Messages Covering Insights Learned During the Tutorial

One thing I want to look into just because my safety relies on this research is whether the three summer months (June, July, and August) are safer on average than the others.

#averaging crime rates per month
avgs <- crimes %>%
  group_by(month) %>%
  summarize(num_crimes = n()) 
avgs

## # A tibble: 12 x 2
##    month num_crimes
##    <chr>      <int>
##  1 01         10165
##  2 02          8300
##  3 03         10320
##  4 04         10986
##  5 05         12700
##  6 06         13495
##  7 07         13217
##  8 08         13001
##  9 09         11942
## 10 10         11903
## 11 11         10421
## 12 12          9671

#we can do some simple computations to compute average manually with these numbers
#average summer crime numbers for Minneapolis
summer_avg <- (13495 + 12317 + 13001) / 3
summer_avg

## [1] 12937.67

#average crime numbers for anything except summer for Minneapolis
rest_avg <- (10165 + 8300 + 10320 + 10986 + 12700 + 11942 + 11903) / 7
rest_avg

## [1] 10902.29

#now let's reduce it to just looking at the University of Minnesota where I'll be living
avgs_umn <- crimes %>%
  filter(Neighborhood == "UNIVERSITY OF MINNESOTA") %>%
  group_by(month) %>%
  summarize(num_crimes = n()) 
avgs_umn

## # A tibble: 12 x 2
##    month num_crimes
##    <chr>      <int>
##  1 01            37
##  2 02            51
##  3 03            46
##  4 04            48
##  5 05            69
##  6 06            57
##  7 07            54
##  8 08            66
##  9 09            79
## 10 10            69
## 11 11            56
## 12 12            31

#average summer crime numbers for UMN
summer_umn <- (57 + 54 + 66) / 3
summer_umn

## [1] 59

#average crime numbers for anything except summer for UMN
rest_umn <- (37 + 52 + 46 + 48 + 69 + 79 + 69) / 7
rest_umn

## [1] 57.14286

This doesn’t look good, nor is it what I was expecting. I didn’t do my research on the city because I just wanted whatever internship I could get but firstly it definitely isn’t any safer than the average for America. Also, the Summer months are specifically more dangerous for both the entire city and the campus.

Let’s look at these numbers in terms of the populations of the city and campus though.

#all of these numbers for population from Google

summer_rate = summer_avg / 413651
summer_rate

## [1] 0.03127677

rest_rate = rest_avg / 413651
rest_rate

## [1] 0.02635624

umnsummer_rate = summer_umn / 51147
umnsummer_rate

## [1] 0.001153538

umnrest_rate = rest_umn / 51147
umnrest_rate

## [1] 0.001117228

Conclusion

I have learned two important things by my data science endeavors. Firstly, where I’ll be living in the University of Minnesota area is much safer than the general city of Minneapolis relative to its population. This could change as population might drop significantly in the summer due to students moving away but I couldn’t find data on that. Secondly, the campus is roughly just as safe as it is during the rest of the year in the summer which is good.

My graph earlier limited data to recent years so the graph would be more concise and succinct but using the data over the years shows that the data points to two things. Minneapolis matches the average crime rate in United States which means it’s not overly safe or dangerous by our hypothesis testing and there is a similar crime count during the summer months.

Further Reading:

Data sets taken from: https://www.kaggle.com/mrisdal/minneapolis-incidents-crime Hypothesis/crime statistics from: http://www.nationmaster.com/country-info/stats/Crime/Total-crimes-per-1000 Further reading and data on how safe Minneapolis-St. Paul, Minnesota is: https://realestate.usnews.com/places/minnesota/minneapolis-st-paul/crime Read more on linear regression and other statistical methods used: http://www-bcf.usc.edu/~gareth/ISL/